Balderton 
- Portfolio News
- 04 March, 2025
Prior Labs has built a state-of-the-art AI model for tabular data to unlock the huge untapped value in spreadsheets and databases used by businesses worldwide.
Published in Nature, the TabPFN model makes learning on small tabular data faster (in seconds, not hours), more accurate and effective with half the data. The team is now turning this academic success into real-world, commercial impact by allowing businesses to integrate TabPFN through self-hosted models, or through an API.
The pre-Seed funding was led by Balderton along with XTX Ventures, SAP founder Hans Werner-Hector’s Hector Foundation, Atlantic Labs, and Galion.exe. Prominent AI angel investors such as Peter Sarlin (Founder & CEO, Silo AI), Thomas Wolf (Founder & CSO, Hugging Face), Guy Podjarny (Founder, Snyk & Tessl), Ed Grefenstette (Director, DeepMind), Robin Rombach (Founder & CEO, Black Forest Labs), Chris Lynch (Founding Investor Data Robot & CEO, AtScale), Ash Kulkarni (CEO, Elastic) and other business leaders also participated. The funding will accelerate product development, expand the team, and bring Prior Labs’ groundbreaking foundation model to more users.
Tabular data is the backbone of science and business, yet the AI revolution transforming text, images and video has had only a marginal impact on tabular data – until now. Prior Labs’ breakthrough gives everyone the super-powers of machine learning without needing to train their own models on their own data. We’re thrilled to support this world-class team as they redefine how industries unlock the value of their data.
James Wise Partner, Balderton
Prior Labs was founded in late 2024 from within the ELLIS ecosystem by Professor Frank Hutter, the world’s most cited AutoML researcher; Noah Hollmann, a computer scientist experienced at Google and BCG and X-Prize finalist; and Sauraj Gambhir, a former venture capital, M&A and enterprise growth expert. Bernhard Schölkopf, a leading AI pioneer (Director at ELLIS & Max Planck Institute Tübingen), and entrepreneur and investor Alex Diehl (Co-Founder of Architizer, KKLD*, and BMW iVentures) are Prior Labs’ founding advisors.
With 20+ years of experience in machine learning, Hutter‘s team leveraged their expertise to create a state-of-the-art foundation model for tabular data. Their groundbreaking work, detailed in a paper recently published in Nature, showcases the transformative potential of TabPFN. Now, Prior Labs is scaling this academic success to deliver real-world impact by integrating their API into business’ data workflows, thereby enabling businesses to unlock the hidden potential of their tabular data.
Bridging the AI gap in structured data
Tabular data—structured data in tables, spreadsheets, and databases—underpins critical operations in healthcare, finance, environmental monitoring, and manufacturing. Despite its importance, tabular data analysis has lagged behind the rapid advances seen in AI for text and images. Challenges such as messy, diverse, and context-specific data have left businesses reliant on outdated tools or costly, bespoke machine learning models for each task.
Prior Labs is driving a paradigm shift. Its groundbreaking TabPFN model offers a universal solution for tabular data analysis. Trained on 130 million synthetic datasets, TabPFN is designed to understand and infer patterns in any dataset instantly, without requiring task-specific training. As a foundation model, it also allows fine-tuning with a company’s proprietary data, continuously improving its accuracy and adaptability to real-world challenges.
Most of the world’s critical decisions are powered by tabular data, yet tools to analyse it are outdated and lacking. We’re bringing a quantum leap to the predictions that businesses can make from their most valuable data and building a future where engaging with tables is as seamless as using AI for text or images. We can deliver faster, more accurate predictions that empower businesses to do more with less.
Frank Hutter co-founder and CEO, Prior Labs
In the recent Nature paper, TabPFN was shown to outperform the accuracy of state-of-the-art models in over 96% of use cases on small tabular data. It requires 50% of the data to reach the same level of accuracy as the next best model and only takes 2.8 seconds to deliver better performance than the best existing models in 4+ hours. It is remarkably easy to use and can be applied to any dataset quickly with just a few lines of code.
In industries like trading, finance, and business analytics, TabPFN enhances profitability by delivering faster, more accurate predictions that inform critical decisions. Similarly, in data-constrained fields such as healthcare, medicine, and climate science, where acquiring data is often challenging and expensive, TabPFN delivers the same high-quality results with 50% less data, opening doors for groundbreaking scientific discoveries.
Rapid advancement
Prior Labs is now expanding on this success with an enhanced API enabling businesses to seamlessly use TabPFN’s capabilities and integrate its benefits into their operations at scale. Prior Labs also continues to rapidly advance the speed, accuracy and efficiency of its baseline model. The latest advancements include support for text features, fine-tuning on proprietary data and the ability to incorporate contextual information about the prediction task further increasing accuracy and ease-of-use. TabPFN can also be used for time-series forecasting where it currently ranks #1 on the industry standard GIFT-Eval benchmark, ahead of Amazon’s popular Chronos model.
